2 ms·
I don't think that's true on transformer models. Ignoring caches+optimisations, a transformer model takes as input a string of words and generates one more wor
by londons_explore 2y ago
I don't think that's true on transformer models.
Ignoring caches+optimisations, a transformer model takes as input a string of words and generates one more word. No other internal state is stored or used for the next word apart from the previous words.
- delusional 2y agoThe words in the disclaimer would have to be the "hidden state". As said, this is unlikely to be true, but theoretically you could imagine a model that starts outputting a disclaimer like "as a large language model" it's possible that the next top 2 words would be "I" or "it" where "I" would lead to correct answers and "it" would lead to wrong ones. Blocking it form outputting "I" would then preclude you from getting to the correct response. This is a rather contrived example, but the "mind" of an AI is different our own. We think inside of our brains and express that in words. We can substitute words without substituting the intent behind them. The AI can't. The words are the literal computation. Different words, different intent.